Vanguard Program
Advanced architecture prototype supercomputers for the U.S. DOE/NNSA
The Vanguard Program is a U.S. Department of Energy / National Nuclear Security Administration (NNSA) initiative to prototype advanced supercomputing architectures ahead of large-scale production system procurements. Vanguard bridges the gap between small-scale hardware testbeds and leadership-class production systems, evaluating novel processor architectures, memory technologies, and system software stacks to reduce deployment risk and inform future NNSA systems under the Advanced Simulation and Computing (ASC) program.
Andrew serves as Technical Advisor for the Vanguard program, supporting architecture evaluation and system software strategy.
Vanguard II: Spectra
Spectra is the second system deployed under the Vanguard program, built in partnership with NextSilicon and Penguin Solutions. Spectra achieved full system acceptance in May 2026.
The system features 64 compute nodes equipped with 128 Maverick-2 accelerators — NextSilicon’s second-generation chips based on the Intelligent Compute Architecture (ICA), a dataflow architecture that allows the chip to dynamically rewire itself to adapt to changing workloads in real time. Each node hosts two dual-die Maverick-2 OAM devices, integrated with OCP-based Tundra servers from Penguin Solutions and a liquid cooling system from Chilldyne.
Rather than treating all compute tasks equally, Maverick-2 chips analyze code to prioritize operations dynamically, targeting significant efficiency gains — up to 10x performance improvement at half the power of conventional approaches.
Sandia has successfully validated Spectra on NNSA ASC workloads, including HPCG, the LAMMPS molecular dynamics simulation, and SPARTA, establishing Spectra as a first-of-its-kind adaptive computing capability for nuclear deterrence simulations.
Vanguard I: Astra
Astra was the world’s first Petascale supercomputer based on the Arm processor architecture, deployed at Sandia National Laboratories in 2018. Built around Cavium ThunderX2 processors across 2,592 compute nodes, Astra achieved over 2.3 petaflops of theoretical peak performance.
Andrew was a key contributor to Astra’s deployment, application readiness, and the Advanced Tri-lab Software Environment (ATSE). Astra demonstrated that Arm-based processors could deliver competitive HPC performance for NNSA workloads, helping pave the way for Arm’s broader adoption in supercomputing — including the Fugaku system, which reached #1 on the TOP500 list. Related publications: [1], [2], [3]